Product
Bid and proposal management software with governed AI.
BidAngel runs a bid from the opportunity to the debrief. AI reads the solicitation and writes first drafts from content your team has already approved. A named person approves what goes forward at each step.
Every claim traces to its evidence and to the person who approved it.
- A draft is written only from library content your team has approved.
- Each claim carries a citation to the item it came from.
- Each cited claim is checked against that source and marked supported, partly supported or unsupported.
- A person accepts, edits or removes every flagged claim before the draft can be approved.
- The approval is recorded with the approver’s name, and the record cannot be edited.
Nine steps, in the order the job is done
What is in the product at each step. How each step runs →
Find the opportunity
Scheduled imports from SAM.gov and AusTender, with each amendment kept as a version. Upload for an RFP you received directly. An inbox where your team watches, reviews or dismisses each opportunity.
Understand the solicitation
Requirements and evaluation criteria extracted from the solicitation documents, each linked to its source passage. Your team approves, edits or rejects each one.
Decide whether to bid
A fit score with its reasons shown, flags for anything that rules a bid out, and a go/no-go decision that only a person can record.
Plan the response
A compliance matrix built from the approved requirements. Tasks with owners and deadlines. A place to record your position on each of the evaluation factors the buyer published.
Reuse what you have proved
A content library searched by meaning and by keyword. Each item has an owner, an approval state and a freshness state. Past questionnaires can be imported as question-and-answer pairs.
Draft
Section drafts with a citation beside each claim. A marked gap where no approved content covers a requirement. A check of each cited claim against its source, marked supported, partly supported or unsupported. One run can draft every section of the response.
Review and approve
Reviewer assignment, a queue of flagged claims to accept, edit or remove, a side-by-side comparison of draft versions, and approval by a proposal lead.
Assemble the submission
A coverage report of mandatory requirements against approved responses, and a versioned submission package of approved items with a record of who approved each one.
Learn from the result
A win/loss record with reason codes, and debrief scores recorded against each evaluation factor.
CRM Integration (HubSpot available on request; Salesforce and Dynamics coming)
HubSpot is available, switched on per workspace on request: import companies and deals, link a deal to an opportunity or pursuit, and write the pursuit's stage and go/no-go status back to the deal. Salesforce and Microsoft Dynamics are coming next.
Bring your own agent
A workspace admin issues a credential from settings with a chosen role, a list of permitted tools, and optional usage and rate limits. Copy the secret once when it is issued and paste it into your agent’s configuration.
Your agent connects to BidAngel’s MCP endpoint and uses the same tools as our own drafting agent. Drafting from approved content only, gap markers, and approval by a person all apply in the same way, because they are enforced by BidAngel and not by the agent. A draft from your agent appears in the same pursuit view, labelled with the agent that wrote it, and is a suggestion until a proposal lead approves it.
Your agent and ours pass the same permission checks, see only your workspace’s data, and leave the same audit trail. The only difference is the agent identity recorded on each action.
How buyers find you, before any RFx is written.
The nine steps above are how you respond to a bid. Buyers also look for suppliers before any RFx exists, and we expect buyer-side AI agents to do more of that looking: building a shortlist, or checking who can prove a capability.
Today you can maintain approved capability claims and list your firm in an opt-in directory. Credentialled queries by buyer agents, with signed responses, are planned and not yet available. There is no marketplace and no scoring by BidAngel. You alone decide what is published.
What is available today, and what is planned →Governed by construction
Every tool call — human, first-party agent, or external agent — runs through one governance pipeline. Class A reads, Class B low-risk writes, and Class C consequential writes are classified at the tool level and enforced at the boundary. Consequential actions require explicit human approval regardless of which caller invoked them.
Credentials are hashed at rest with argon2id, verified in constant time, revocable with overlap-based rotation, and scoped to a bound role and tool allowlist frozen at issuance. Raw secrets are shown exactly once in the admin UI and are never logged, echoed in errors, or stored in audit payloads.
What BidAngel does not do
- It does not submit to buyer portals. BidAngel assembles the package. You submit it.
- It has no capture plan, no price-to-win model and no competitor analysis of its own. You can record the competitors you expect on each pursuit, with the incumbent and their likely strengths and weaknesses. It is strongest from the compliance matrix onwards.
- It holds no third-party security certification yet. We are working toward SOC 2. The security page says where that stands.